Optimization:

TRANSYT-7F is capable of optimizing CORSIM networks with pre-timed and semi-actuated signals, where CORSIM simulates the candidate timing plans, and TRANSYT-7F manages the genetic algorithm.  Release 11 (targeted for March 2007) can directly optimize fully-actuated signals in CORSIM, and directly optimize multi-period CORSIM networks.

The optimization node list is mentioned in the documentation, but is probably not emphasized enough.  For large networks, or for direct CORSIM optimization, it is very helpful to remove non-critical intersections (e.g., already operating at LOS A or B) from the node list, for more efficient optimization.

The Original Cycle Length (OCL) affects the first individual in the first generation of genetic algorithm optimization. The first individual contributes to the optimization process, and is also used to create "Initial" results for the output file. Typically the first individual in the first generation of optimization simply reflects the user-coded initial timing plan. However, if the initial timing flag is turned on, coordinated nodes on the optimization node list immediately adopt the OCL. All other individuals in the first generation are unaffected by the OCL. Subsequent individuals in subsequent generations aren't explicitly affected by the OCL. However, they are usually implicitly affected by gene pool contributions from the first individual in the first generation.

TRANSYT-7F provides CORSIM processing and optimization for TSIS-CORSIM versions 5.1 and 6.0.

Traditional TRANSYT optimization takes into account the reduced traffic flows due to queue spillback, unless the chosen performance index (PI) is PROS-only. All other PI's (besides PROS-only) consider delay, or throughput, or both. When maximizing throughput, the timing plan that maximizes the (reduced) traffic flows is recommended as optimal. When minimizing delay, the timing plan that maximizes the (reduced) capacity of each movement is recommended as optimal, because lower capacities produce higher delay estimates.

Recent versions of TRANSYT-7F are capable of optimizing cycle lengths at uncoordinated intersections within the network, independent of the network background cycle length. It is also possible to optimize phasing sequence and splits at uncoordinated intersections. However, since phasing sequence optimization is mostly beneficial for improving progression, it is not always beneficial for uncoordinated intersections.

The initial timing model, which corresponds with the initial timing "flag," is useful for instantly generating a reasonable and effective set of green times from scratch. Users typically request initial timing from TRANSYT-7F when the existing timing plan is unknown, or when they think the program may be able to develop an effective starting point for optimization. The initial timing model allocates green time in an attempt to equalize the degree of saturation on each movement. This quick estimation technique is rarely sufficient for finding the global optimum timing plan, but does provide a reasonable and effective set of initial green times. Initial timing can only be provided for certain intersections when 1) the intersection possesses a "coordinated signal," 2) the intersection is present on the "optimization node list," and 3) the initial timing "flag" has been turned on.

Genetic algorithm optimization runs can locate the optimum solution more quickly when "unimportant" signal settings are omitted from the optimization process. By ignoring (i.e., holding constant) the signal settings that are likely to have a negligible effect on results, the program is able to examine more combinations of the signal settings that have a more significant effect on results.

For example, when traffic networks are significantly undersaturated, such as having 20 or fewer seconds of delay per vehicle, optimization of offsets and phasing sequence tends to have a negligible effect on results. Similarly, offsets and phasing sequence are likely to have little impact within severely oversaturated networks, in which there are very few opportunities for progression. Finally, offsets and phasing sequence are likely to have little impact at isolated intersections with no nearby upstream signal. In cases such as this, genetic algorithm optimization runs can locate the optimum solution more quickly when offsets and phasing sequence are omitted from the process, and only the cycle length and splits are optimized.

For a second example, when traffic networks are near capacity and have numerous progression opportunities, optimization of offsets and phasing sequence tends to allow significant improvements in flow and performance. In cases such as this, genetic algorithm optimization runs can locate the optimum solution more quickly when cycle length and splits are omitted from the process, and only offsets and phasing sequence are optimized. For this type of optimization, it is necessary to specify the appropriate cycle length in advance, and the software can use the initial timing model to estimate new splits for every candidate phasing sequence.

Release 9.4 introduced phasing sequence optimization, using the genetic algorithm. Phasing sequence can be optimized on both the major and minor streets. During optimization, TRANSYT-7F examines virtually all feasible phasing sequences including leading and lagging left-turns with and without overlap, lead-lag phasing, and split phasing. Relative to right-hand driving, the program is designed to be equally effective in optimizing phasing sequences for networks with left-hand driving (e.g., having leading or lagging right-turns). The user may allow full optimization throughout the network, but can also specify restrictions on optimization at any approach of any intersection. On every approach, the user may specify whether they wish to allow optimization, overlap phasing, lead-lag phasing, leading through movements, or the “yellow trap”.

The genetic algorithm optimization module implements a default population size of 10 individuals. This means that 10 simulation runs are performed (each with a unique signal timing plan) for each generation, prior to crossover and mutation. Release 9.6 was the first to allow the user to modify the population size, which sometimes allows the genetic algorithm to locate the global optimum solution for certain networks in a shorter amount of time. The genetic algorithm optimization module from HCS (called SOAP2K) also implements a default population size of 10 individuals.

For optimization runs it is necessary to specify an optimization node list within the input file. Signal timing is not optimized for intersections that are omitted from this list. There is no required sequence of node numbers. When hill-climb optimization is used, changing the optimization node sequence can sometimes result in better signal timing plans generated by the program. In addition, nodes can be specified multiple times on the node list if they deserve special attention, although this may increase program running times on the computer.

Numerous objective functions are available to specifically take aim at delay, stops, queuing, fuel consumption, progression, throughput, or combinations of these. Also, TRANSYT-7F can apply global, node-specific, and link-specific optimization weighting factors.

The traditional objective functions (DI and PROS) involve measurements of delay, stops, fuel consumption, and progression opportunities. However, under severely oversaturated conditions, measurements of delay, stops, and fuel consumption are known to become less accurate, thus compromising the effectiveness of the DI objective function. Similarly, under severely oversaturated conditions, increases in progression opportunities may not address spillback problems, thus compromising the effectiveness of the PROS objective function. When traditional optimization strategies fail, involving old objective functions (DI, PROS, PROS/DI, PROS & DI), queue & stop penalties, link & node penalties, etc., the new objective functions may provide superior signal timing for severely oversaturated conditions. The new objective functions involve throughput and queuing ratio.

TRANSYT-7F is capable of supplying a reasonable and effective initial timing plan prior to optimization. If the user requests for initial timing to be generated by the program, it is only necessary for them to supply the fixed yellow and all red intervals, and then the program supplies the green intervals. The technique that is applied to generate the initial green intervals at a pre-timed signal is similar to the HCM planning analysis methodology. If any phases are flagged as actuated, then preference is shifted to the non-actuated phase(s) in the allocation of green within the initial timing model. However, if the user has an alternative timing plan that is reasonable and effective, they are encouraged to try it out also because the initial splits and offsets that are specified can have a significant impact on the hill climb optimization search process.

Simulation:

TRANSYT-7F now contains two models for traffic-actuated control. The original target degree of saturation model, employed by TRANSYT-7F for many years, is effective for optimization runs. This model facilitates location of the optimal actuated phase times. The newer actuated signal timing estimation module is effective for simulation runs, and evaluation of existing conditions. The estimation module computes traffic-actuated phase times that are likely to materialize in the field, based on the user-specified maximum green and force-off settings.

If lane utilization is highly unequal on a multilane lane group, this can be simulated effectively by defining a link for each lane. Up to 50 links, and as many approaches as needed, can be simulated at any given intersection. This is why TRANSYT-7F is extremely flexible for modeling highly unusual intersection geometries.

The degree of platoon dispersion on internal links can be calibrated for local conditions by using the platoon dispersion factor (PDF). High platoon dispersion factors indicate heavy friction, i.e. urban CBD areas having significant amounts of parking, turning, pedestrians, and narrow lane widths, which conspire to reduce platoon intensities. Low platoon dispersion factors indicate low friction, i.e. ideal suburban high-type arterial street conditions that allow increased platoon intensities.

Queue spillover and queue spillback only occur on internal links, or external links that are sufficiently long. The concept is that if an external link is coded with a short link length, then spillback on that link is not supposed to affect the optimization objective function.

As with CORSIM, user-defined link lengths cannot be too short. Link lengths and free flow speeds must not allow a vehicle to completely skip over any link during one second of simulation. If a vehicle is capable of skipping over a link completely, this can potentially compromise simulation results or cause fatal errors.

Output:

Release 8.2 (in 1998) was the first version ever to compute control delay. Prior versions offered stop delay and/or total delay. All releases since 8.2 have provided control delay estimates. Delay is computed by the same methods from the HCM2000, but augmented by macroscopic simulation results. Values of capacity and the percentage of vehicles arriving on green (PVG) are obtained from simulation, instead of user input.

TRANSYT-7F computes the fuel consumption due to random-plus-oversaturation stops differently than the fuel consumption due to uniform stops. The model assumes that random-plus-oversaturation stops are made while vehicles are traveling at the queue speed (the one at which the queue dissipates), while uniform stops are assumed to be made from cruise speed. The queue speed is calculated as follows: Queue Speed = (Saturation Flow / Queue Density).

TRANSYT-7F can report arterial performance statistics if route summary reports are requested within the input file, and if one or more progression routes have been defined within the input file. Some of the sample input and output files on the McTrans web site demonstrate this.

The wide, 132-column output format contains some useful results that are not displayed in the narrow, 80-column output format. For example, an output parameter called Flow indicates the number of vehicles discharged during simulation. For a given link, if Flow is lower than the input volume, this may indicate inadequate capacity due to spillback, or possibly insufficient green time.

Efficiency is defined as the shortest major street green divided by the bandwidth in each individual direction. If efficiency is 50%, this implies that progression has been maximized, because the bandwidth in each direction uses the entire available major street green. Attainability is defined as bandwidth divided by the shortest major street green. If attainability is 100%, this implies that the bandwidth cannot be increased without increasing the shortest green time available along the major street. Efficiency and attainability are only reported within the TRANSYT-7F output file when a specific route has been defined in the input file.

TRANSYT-7F's computation of uniform delay (d1) can be confirmed through use of the Spyglass utility program. Using Spyglass, the uniform queue length from simulation can be plotted within spreadsheet software. Subsequently, the uniform delay value can be obtained by computing the area under the uniform queue profile. When oversaturated conditions prevail, residual demand delay (d3) is included within the value of uniform delay. The computation of incremental or random delay (d2) can be confirmed by using a calculator or spreadsheet to apply the Highway Capacity Manual formula for incremental delay.

Input:

There are two ways to model an exclusive pedestrian phase in TRANSYT.  The first method is to code an empty phase with no moving links.  The second method is to code a separate phase with a low-volume (e.g., 10-vph) traffic movement, using an extra link (e.g., link 113).  The optimization process will likely force such an exclusive pedestrian phase length to its minimum specified duration, especially if no pedestrian links with pedestrian traffic are modeled.  Because of this, the minimum phase length specified should be sufficiently long to ensure that the variable green interval is not reduced to an unsafe length.

"Mid-block source volume" is a significant input parameter in TRANSYT-7F, because non-progressed flow is not assumed to be uniform flow. Instead, progressed feeding flows will be factored up or down by the traffic model to exactly match the downstream flow, in the absence of any coded mid-block source volume. This differs from "external links", at which all volume is assumed to be uniform flow.

Offsets and yield points are referenced to the “reference interval”. To achieve consistency with CORSIM or controllers in the field, it is usually preferable to define the reference interval as the first fixed interval (i.e., flashing don't walk, or yellow) following the major street's green interval.